Re: [Xen-devel] [PATCH v8 3/5] xen: don't process rcu callbacks when holding a rcu_read_lock()

["Jürgen Groß" <jgross@suse.com>]

On 27.03.20 09:35, Jan Beulich wrote:
> On 27.03.2020 09:10, Jürgen Groß wrote:
>> On 27.03.20 00:24, Igor Druzhinin wrote:
>>> On 26/03/2020 09:19, Juergen Gross wrote:
>>>> Some keyhandlers are calling process_pending_softirqs() while holding
>>>> a rcu_read_lock(). This is wrong, as process_pending_softirqs() might
>>>> activate rcu calls which should not happen inside a rcu_read_lock().
>>>>
>>>> For that purpose modify process_pending_softirqs() to not allow rcu
>>>> callback processing when a rcu_read_lock() is being held.
>>>>
>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>> Reviewed-by: Jan Beulich <jbeulich@suse.com>
>>>> ---
>>>> V3:
>>>> - add RCU_SOFTIRQ to ignore in process_pending_softirqs_norcu()
>>>>     (Roger Pau Monné)
>>>>
>>>> V5:
>>>> - block rcu processing depending on rch_read_lock() being held or not
>>>>     (Jan Beulich)
>>>
>>> Juergen,
>>>
>>> Our BVT revealed a likely problem with this commit in that form.
>>> Since 12509bbeb9e ("rwlocks: call preempt_disable() when taking a rwlock")
>>> preemption is disabled after taking cpu_maps which will block RCU
>>> callback processing inside rcu_barrier itself. This will result in
>>
>> Why would that block RCU callback processing?
>>
>> RCU callbacks should be blocked only if a rcu lock is being held.
>>
>> Did I miss something in my patches?
> 
> Igor, are you perhaps running without "rcu: add assertions to debug
> build"? I think this actually fixes what you describe. Without it
> rcu_barrier(), in its second loop, calling process_pending_softirqs(),
> would cause the RCU softirq to not be invoked anymore with preemption
> disabled. Of course the title of this change doesn't reflect this at
> all.

Right. This explains why I don't see the hang on my test system.

> 
> Jürgen, as an aside, while looking at the code again, I think the
> comment near the end of process_pending_softirqs() would now rather
> belong at its very beginning; should have spotted this while
> reviewing.

Oh, indeed. Will send a patch.


Juergen